How does market access shape internal migration?∗
Laura Hering
†
Rodrigo Paillacar‡
May 5th, 2012
Abstract
This paper studies the role of international trade on internal migration in
an economic geography framework by assessing the impact of regional differences in access to foreign markets on bilateral migration rates between the 27
Brazilian states. We use individual data for 1993 to 2003 to construct yearly
migration rates. Bilateral migration rates are also computed separately for
each sector and for different educational levels. State and sector-state specific market access indicators are estimated using trade data. We find that
regions with low market access see residents migrate to regions with higher
market access, where higher labor demand leads to better job opportunities.
Low-educated migrants are found to be more sensitive to regional differences
in market access than high-educated. This could be explained by the specialization of Brazil in goods that are intensive in unskilled labor. An increase in
the demand for Brazilian goods thus leads to a higher demand for low educated labor. All manufacturing workers react strongly to regional differences
in market access of their own industry.
Keywords: Economic geography, migration, international trade, wage, selection bias, Brazil.
JEL classification: F12, F16, R12, R23.
∗
We thank Andrew Bernard, Maarten Bosker, Matthieu Couttenier, Fabian Gouret, Philippe
Martin, Sandra Poncet, Loriane Py, Vincent Rebeyrol, Jose de Souza, Cristina Terra, Julien Vauday, Vincent Vicard and participants of the ELSNIT conference (Florence), the Migration and
Development Conference (Lille), the CREST Macro seminar (Paris) and the EUI Macro seminar
(Florence) for excellent suggestions and fruitful discussions.
†
Erasmus School of Economics. Department of Economics P.O. Box 1738, 3000 DR Rotterdam,
The Netherlands. E-mail: laura.hering@gmail.com
‡
University de Cergy-Pontoise, Laboratoire THEMA, 33 Boulevard du Port, 95011 CergyPontoise Cedex, France. E-mail: rodrigo.paillacar@u-cergy.fr
1
1
Introduction
Nearly everywhere in the world, globalization has led to a substantial and durable
change in domestic labor market conditions. Opening a country to international
trade is likely to have different impacts across regions and sectors. Depending on
the industry or region under study, trade has been found to drive wages are either
up or down, to create, to change or to destroy jobs, affecting by this the spatial
distribution of economic activity within the country (OECD report 2009).
The aim of this paper is to emphasize the role of international trade on the
domestic migration pattern by studying its impact on inter-state migration within
Brazil.
Even though the link between migration and trade has been at the center of
many studies, these have mainly focused on the reverse causality, the impact of
international migration on shaping international trade1 . Yet, in an open economy
trade should affect also internal migration patterns. In fact, internal migration flows
are of a much bigger magnitude than international flows and worker immobility
across regions can explain in great part negative impacts of trade openness.
A theoretical framework that is particular well suited to study the impact of
trade on the spatial structure of a country, taking into account heterogeneity in
sectors, is the New Economic Geography (NEG) theory. This theory explains the
agglomeration of economic activity and how workers will choose their location following changes in trade costs. It underlines the importance of the region’s proximity
to consumers, which is modeled by the region’s access to markets. Regions closer to
big consumer markets (i.e. with higher “market access”) experience lower transport
costs and can sell at lower prices in these markets. Firms located in high market
access regions thus face higher demand for their products and consequently higher
labor demand. Another important feature of high market access regions is that they
are more likely to pay higher wages because of the low transport costs, resulting in
higher income levels. This effect is reinforced since the costs of living in these regions
are often lower because also imported consumption goods incur lower transport costs
(Krugman, 1991).
This paper investigates how a country’s integration into the world economy influences spatial labor mobility within the domestic economy. Sectors that enjoy a
comparative advantage on the international market will see the demand for their
products grow. As a consequence, firms in these sectors will be able to pay higher
wages or recruit more workers. Firms doing business in the advantaged sectors and
located in regions that are closer to the new markets, will benefit even stronger from
the new demand. Since they have the advantage of lower transport costs in serving
the new destinations, they will see the demand for their products increase more than
firms in more remote areas and thus attract more migrants.
Here, we show how internal migration patterns within Brazil are influenced by
the country’s access to international markets using a panel data set for the years
1995 - 2003. Bilateral Migration rates between the 27 Brazilian states are computed
separately for twelve sectors. Regional differences in trade performance are captured
1
Studies on the impact of immigration on trade are numerous. See for example Gould (1994),
Head and Ries (1998), Rauch and Trindade (2002). Combes et al. (2005) look at the impact of
internal migration on internal trade.
2
by a structural measure of each region’s access to foreign consumer markets. This
market access indicator is constructed using actual trade data and gives us the
export capacity of a certain state for a given sector. In the regression analysis we
then show how the migration decision of workers is influenced by regional differences
in market access. Undertaking our analysis at the sectoral level allows us to assess
the impact of a region-sector specific market access on migrant flows in each sector
separately and to test for possible heterogeneity across industries. That means, we
test whether the migration decision is driven by the evolution of market access in
the particular industry of the migrant.
Two features make Brazil a particular interesting case to study.
First, the country experienced an important trade integration in the beginning
of the 1990s. In 1991, Brazil entered into negotiations about the creation of the free
trade agreement Mercosur with its neighbors Argentina, Uruguay and Paraguay.
The establishment of this free trade zone in 1995 had differentiated effects across
states and sectors (source).
Second, Brazil exhibits exceptionally high levels of internal migration: in 1999,
40% of the all Brazilians lived in a state different from the state where they were
born (Fiess and Verner, 2003).2
However, Menezes-Filho and Muendler (2007) find that the high migration rate
in Brazil stands in sharp contrast to the low sector relocation in the country. Further,
recent empirical work by Muendler et al. (2011) has highlighted that even in the
presence of structural changes of industries resulting from their competitiveness on
international markets, workers rarely change sectors. These authors highlight that
when losing their job, workers are more likely to search a job in the same sector,
often in the informal economy, or leave manufacturing industries all together and
go into the (often informal) service sectors. Despite the high migration rate within
the country, the alternative strategy, spatial reallocation within sectors, has so far
received little attention by researchers.
In this paper, we argue that regional differences in international trade performance are likely to play a significant role on migration decisions for several reasons.
A region which increases its access to foreign consumer markets attracts firms and
raises labor demand. Besides the direct effect on spot wages (which could be dissipated following a high increase of immigration), volatility of workers’ income over
the long run could be reduced if high market access regions facilitate labor matching
and host more productive firms. Even for workers in sectors that are suffering from
import competition, regions with a relatively high market access could provide a
(temporary) shelter for those workers unable to switch to sectors.
In our analysis, we combine spatial relocation with comparative advantage considerations by focusing on migration responses at the sectoral level. This means, we
test whether the migration decision is driven by regional changes of international
market access in the particular industry of the worker. Moreover, although we also
study the impact of local demand, our main focus in the paper is the region’s access
to international markets. We exploit the variation over time of the market access
differentials in a panel regression setting, where we control for sector-year specific
2
For a comparison: Dahl (2002) finds that in 1999 in the US, a country known for its high
labor mobility, only 30% of all male workers no longer live in the state in which they were born.
Kovak (2011) reports an interstate mobility of 29% using information from the Brazilian Census.
3
wage differentials and other time-varying determinants of bilateral migration like
regional differences in unemployment and population among others.
This paper contributes to the literature in at least four ways. We are the first to
highlight the link between trade and internal migration in an economic geography
framework. While Crozet (2004) concentrates on the importance of low living costs
generated by a good domestic market access, we concentrate on the role of the access
to foreign markets.
Second, by undertaking our analysis at the sectoral level and using only access
to international markets and not to domestic markets has several important advantages. First, we eliminate an endogeneity problem that arises when immigrants
raise local consumption, as highlighted by Head and Mayer (2004). Since we are
interested in the role of trade, our market access is constructed using revealed trade
flows, as do Redding and Venables (2004). The use of bilateral trade data between
the Brazilian states and the foreign countries reveals both observed and unobserved
trade costs and market characteristics, both for foreign countries and for the Brazilian states. These allow us to obtain a structural measure of market access. The
availability of trade flows at the sectoral level allows us to undertake our analysis
at the sectoral level. Next to the reducing the endogeneity problem, sectoral market access is likely to be more relevant for the migration decision of the labor force
than the overall market access of a region, since the worker is mostly interested in
the economic opportunities in the sector in which he has already some experience.
Differences in the ranking of regions in terms of market access across industries is
thus also more likely to explain migration flows in opposite directions. A further
important advantage of this sector-specific measure is that it allows us to include
state-time fixed effects in our regressions and thus take away many unobserved state
and time varying factors that could be correlated to the state’s overall market access
(i.e. amenities, price levels, rents).
The third contribution is that we are the first to draw on individual data to
demonstrate the link between economic geography and migration. Using bilateral
migration rates that are obtained from micro data has several advantages. First,
it allows the disaggregation of migration flows and to compare different types of
migrants. Previous studies looking at the impact of economic geography on migration have not done this before. Here, we focus on differences in migration behavior
across two educational levels and 12 industries. Studying the differentiated effects
on skilled and unskilled workers is of particular interest, given that the literature
suggests that the latter should be less sensible to amenities and react more to economic factors like market access. Another reason to focus on unskilled workers is
that Brazil is a relatively unskilled-abundant country (Muriel and Terra, 2009) and
hence these workers should benefit relatively more from trade openness.
Second, it allows us to take into account the migration literature, which highlights
the aspect of a possible self-selection of migrants towards certain destinations. In
our estimation of regional wage levels, we control for a possible self-selection bias in
this variable, as proposed by Dahl (2002). Even though we are looking at bilateral
migration rates at the sectoral level, thanks to the use of micro data, our empirical
strategy is based on the individual migration decision. Our estimated migration
equation is derived directly from the utility differential approach that is widely used
in the migration literature.
4
A last novelty of this paper is that it studies the link between migration and
economic geography in a developing country. Redding and Schott (2003) show in a
theoretical model that if skill-intensive sectors have higher trade costs or stronger
increasing returns to scale, highly skilled workers benefit more from a higher market
access. In line with this model Breinlich (2006) shows on a panel of European
regions, which are specialized in technology and high-skill intensive goods, that
higher market access leads to a stronger accumulation of human capital. Contrary
to the countries of the European Union, studied also by Crozet (2004), Brazil is
highly specialized in goods that are intensive in low-skilled labor. we expect a lower
impact of human capital accumulation through a high market access, because a
higher market access is more likely to increase demand for low-educated workers.
An increase in the demand for these goods is thus likely to affect less the highlyeducated than the low-educated workers. Further, it is possible that migration
decisions of low-killed workers are taken based on different criteria than the decision
of the in general more mobile educated workers. Our second objective is therefore to investigate whether the sensitivity to market access varies across different
educational levels of workers.
Our empirical findings indicate that migration patterns are indeed driven by
regional differences in access to international markets. Our results, obtained from
panel data regressions indicate that workers indeed migrate from states with low to
states with high market access in their respective sector.The effect is stronger for
the migration decision of low skilled labor in sectors with comparative advantage.
We thus find no evidence for the Redding and Schott (2003) prediction. Access to
international markets doesn’t seem to be a key migration determinant for workers
in sectors without comparative advantage and the highly-educated labor force.
We check whether our findings of a positive impact of market access differentials
on migration is robust to the introduction of various control variables relating to
alternative explanations of migration and control for self-selected migration and the
presence of zero-value migration flows.
We finally conclude that altogether, these results suggest that the geography of
regional access to markets can help explain migration patterns within a country. The
fact that migration patterns are affected by the location’s industrial specialization
suggests that implications of the New Economic Geography theory (i.e. regional
advantages generated by the region’s position in the spatial economy) are better
understood in combination with comparative advantage and sector-specific inputs
(e.g. human capital specificity).
The two studies treating the impact of international trade on internal migration
we are aware of are Aguayo-Tellez et al. (2008) and Kovak (2011). As this paper,
both studies apply to Brazil. Aguayo-Tellez et al. (2008) analyze one particular
aspect of exports by looking at the role of multinational enterprises. They find that
migrants are attracted by the growth of employment opportunities in states with
high concentration of foreign owned establishments. Kovak (2011) highlights the
negative effect of imports on labor demand. He shows that decreases in regional
wage levels resulting from tariff cuts will trigger migration between the Brazilian
states.
The fact that previous work on migration determinants has seldom focused on
the trade channel, is partly due to the fact that studies concentrate on labor reloca5
tion rather than on migration per se. Using such an approach, researchers explore
changes in the employment levels, implicitly assuming that factors are mobile. In
the context of NEG models, the effect of market access on employment has been
estimated by Hanson (1998) for Mexico and Head and Mayer (2006) for the European Union. The first study looking at the impact of market access on migration
in a NEG framework is Crozet (2004). For five European countries, he estimates
a quasi-structural migration equation based on the model proposed by Helpman
(1998) to show that cost advantages due to higher market access shape migration
flows within countries. His approach has been adopted by Pons et al. (2007) to explain migration flows within Spain and by Kancs (2005) for migration flows between
the Baltic states.3
The rest of the paper will proceed as follows: Section 2 presents the theoretical
framework, and summarizes some implications for the empirical part; Section 3
outlines the estimation strategy; Section 4 indicates the data sources and describes
the computation of our market access variables and how we obtain consistent regionsector specific wage levels. It also gives some descriptive statistics and stylized facts
on migration and its link to market access in Brazil. Finally, Section 5 gives the
empirical results of the migration equation and Section 6 concludes.
2
Theoretical background
We consider the standard New Economic Geography (NEG) model (Fujita et al.,
1999) that combines monopolistic competition with product differentiation, trade
costs and increasing returns to scale at the firm level. The economy is composed of
j = 1, ..., J regions and one agricultural and s manufacturing sectors. The agricultural sector produces a homogeneous agricultural good A, under constant returns
and perfect competition, which is freely tradeable. Each manufacturing sector s produces a large variety of differentiated goods, under increasing returns and imperfect
competition, where each firm produces a different variety. In both sectors, labor is
the only factor of production, but labor is immobile between sectors.
For simplicity, in this section, we regroup all manufacturing sectors s in one
aggregated manufacturing sector M . This allows us to follow the notation of Fujita
et al. (1999) in the presentation of the model. We will differentiate between sectors
and introduce the superscript s only at the end of this section.
All consumers of region j share the same Cobb-Douglas preferences for the consumption of the goods A and M :
Uj = Mjµ A1−µ
, 0 < µ < 1,
(1)
j
where µ denotes the expenditure share of manufactured goods. Mj is defined by a
constant-elasticity-of-substitution (CES) sub-utility function of ni varieties:
3
There is also a long tradition of economic models treating the connection between industrialization processes and rural- urban migration in developing countries, backing at least at Lewis
(1954) and Harris and Todaro (1970). Our point is that this literature is not addressing the specific
features of a trade channel as NEG models do. In addition, these approaches focus on unilateral
rural-urban migration and are thus not suited when migration flows in opposite directions are
observed.
6
Mj =
R X
(σ−1)/σ
ni qij
σ/(σ−1)
,
σ > 1,
(2)
i=1
with qij representing demand by consumers in region j for a variety produced in
region i and σ being the elasticity of substitution. Given the expenditure on manufacturing goods (M ) of region j (Ej ) and the c.i.f price of a variety produced in i
and sold in j (pij ), the standard two-stage budgeting procedure yields the following
CES demand qij :
σ−1
Ej ,
(3)
qij = p−σ
ij Gj
where Gj is the CES perfect price index for manufactured goods, defined over the
c.i.f. prices:
" R
#1/1−σ
X
1−σ
Gj =
ni pij
.
(4)
i=1
This price index is a (inverse) measure of competition prevailing on market j.
Competition increases with the number of firms active in j (decrease in Gj ) and
decreases with the price at which they sell (increase in Gj ). Each shipment of a
manufactured good from i to j involves a variable “iceberg” transport cost Tij . This
technology assumes that for every unit of good shipped abroad, only a fraction ( T1ij )
arrives. This variable is often proxied by the inverse of the bilateral distance between
i and j. The consumer price pij is then proportional to the mill price pi and Tij . The
demand for a variety produced in i and sold in j Equation (3) can thus be written
as:
qij = (pi Tij )−σ Gσ−1
Ej .
(5)
j
Total sales of a of the representative firm in region i can be obtained by summing
sales across regions. Given that total shipments to one region are Tij times quantities
consumed, this yields:
J
X
pij qij =
j=1
J
X
(pi Tij )1−σ Gσ−1
Ej = pi1−σ M Ai ,
j
(6)
j=1
where
M Ai =
J
X
Tij1−σ Gσ−1
Ej
j
=
j=1
J
X
φij mj ,
(7)
j=1
represents the market access of each exporting region i (Fujita et al., 1999). This
market access indicator is simply the sum of the market capacities of all destinations
j, mj , weighted by the measure of bilateral trade costs, φij , between i and j.
In order to see the different ways how changes in market access can impact the
local economy, we have to model the manufacturing firm’s profit function.
We assume that the technology is the same for all varieties of sector M in all
locations. Each firm faces a fixed input F and a marginal input requirement c.
Given that labor is the only production factor, the quantity of labor li required for
the production of quantity qi can be described as
li = F + cqi .
7
(8)
The firm’s profit function can then be written as
πi = pi qi − wi (F + cqi ),
where
qi =
R
X
(pi Tij )−σ Gσ−1
Ej Tij = p−σ
j
i M Ai ,
(9)
(10)
j=1
Maximizing this profit function using Equation (10) yields the usual mark-up
pricing rule
1
σ
) = cwi ⇔ pi = cw
.
σ
σ−1
Given the pricing rule, the profits of a firm in location i are given by
pi (1 −
π = wi [
qi c
− F ].
σ−1
(11)
(12)
The model supposes free entry and exit of firms in response to profits. According
to the zero-profit condition, the equilibrium output q ∗ is the same for all firms:
q ∗ = F (σ − 1)/c.
(13)
Replacing q ∗ in the labor demand function Equation (8), yields the equilibrium
labor input l∗ = F σ. With Li the number of workers in sector M in region i, the
equilibrium number of manufacturing firms located in i is
ni = Li /l∗ = Li /F σ
(14)
and total output of region i is ni q ∗ .
From this definition we can see that when a region i increases total output, in
the long run, the number of firms and workers in i will have to increase as well.
Under the condition that the free entry of firms leads to zero-profits in the long
run, the equilibrium output q ∗ produced by each firm has to be equal to the demand
function defined in Equation (10).
Each firm maximizes gross profits πij = pij qij /σ on each market. Together with
10 and 11, this yields
"
#
1−σ
J
X
σ
1
(wi ci )1−σ
φij Gσ−1
Ej − wi Fi
πi =
j
σ
σ−1
j
(15)
"
#
1−σ
1
σ
=
(wi ci )1−σ M Ai − wi Fi
σ
σ−1
Setting long run profits equal to zero, using Equation (13) and solving Equation
(15) for wages, we obtain the so-called wage equation (Fujita et al. 1999):
" J
#1/σ
σ−1 X c
i
wi =
φij Gσ−1
Ej
j
σci
F (σ − 1)
j
(16)
σ−1 1
=
[ M Ai ]1/σ .
σc q ∗
8
which shows how wages are influenced by market access. Equation (16) gives
the manufacturing wage at which firms in location i break even, when income levels,
price indices and transport costs are given. It can be seen from this equation that
the wage is higher the higher are the expenditures in the different markets, the
better is the firm’s access to these markets (the lower the transport costs) and the
less competition the firm faces in these markets. Using this equilibrium equation
and replacing the price index by Equation (4) and n by Equatiom (14), we can see
how market access, wages and the number of workers in a given region are linked.
" R
# σ−1
1−σ
R
σ − 1 1 X 1−σ X
1−σ
)[
T
ni pij
Ej ]1/σ
wi = (
σc q ∗ j=1 ij
i=1
σ−1
"
# 1−σ
R
R
X
X
σ−1 1
Li 1−σ
=(
Ej ]1/σ
)[ ∗
Tij1−σ
p
∗ ij
σc q j=1
l
i=1
(17)
In the 90’s, Brazil experienced with the creation of Mercosur and several unilateral tariff reductions an important decrease of trade costs, T , in a high number of
manufacturing sectors. Many Brazilian firms thus enjoyed a gain in market access
due to the liberalization of international trade.
According to the theory above, this trade liberalization should result in the creation of new firms and labor reallocation between locations, with workers going to
locations that experienced an increase in market access. Given that spatial relocation is very slow, we state that under the period of our study (1993 to 2003), Brazil
is still in the phase of adjustment to the lower trade costs.
As can be seen form Equation (17), either wages in region i or the number of
firms and workers in region i will have to adjust, for location i to come back to an
equilibrium situation.
The adjustment by wages occurs when the production of goods do not increase
albeit lower transport costs. Lower transport costs lead to lower consumer prices
which in turn results in an increased demand. But since the additional demand
cannot be satisfied, prices will raise. Firms transfer this gain to workers by paying
higher wages. If the total production of region i stays constant, the increase in the
region’s market access, will result in a proportional increase in the wage level. An
increase in the demand, which is not accompanied by an increase in the production
will thus lead to spatial price (wage) disparities. This is what Head and Mayer
(2006) call the price version of the adjustment to a change in market access.
In the case of the quantity adjustment, the total output of the region increases,
resulting in a higher labor demand Li . New workers are attracted to region i.
These newly arriving workers exert a downward pressure on wages, which (at least
partially) offsets the upward pressure coming from the higher market access (as
described above in the price version). This quantity adjustment via an increase in
ni and Li is reflected in a decrease of the price index of region i, as can be seen from
equation (4).4 The decrease in T is thus compensated by an increase in L.
4
If wages stay constant, the decrease in the price index will have to compensate all of the
original increase to come back to equilibrium.
9
Of course, both adjustment mechanisms can also occur simultaneously. When
labor is not perfectly elastic, some locations with a higher demand for their manufactured goods may pay a higher nominal wage even if the number of firms increases.
Fujita et al. (1999) highlight that locations with a higher demand for manufactures tend, other things being equal, to offer a higher real wage to manufacturing
workers.5 This effect creates a strong agglomeration force, resulting in the concentration of economic activity in certain regions. An increase on market access
attracts migration thus by two channels. The direct impact comes from the creation
of new positions due to higher labor demand or better career opportunities linked to
the firm’s new export activities. Market access also indirectly influences migration
through its impact on wages, which are known to be main drivers of migration.
In this paper, we are interested in the market access indicator of state i in
sector s, M Ais . Does an increase in the difference in market access between state
j, M Ajs and state i, M Ais attracts new workers from i to j? The bigger difference
in market access between the states should result in a relatively stronger increase in
the number of open positions and in the number of export-specific jobs in j (either
because firms grow or because new firms open). Can we see that this contributes to
a relative increase in Ljs in comparison to Lis ? In order to control for the indirect
impact of market access, we also include the regional difference in the sectoral wage
level.
By taking into account the number of new positions but capturing also better
career opportunities, we should capture two different types of migrants. Contracted
migration, including migrants that move in response to a concrete job offer, as well
as speculative migration, where individuals move to a certain region in the hope of
finding a job there, should be responsive to changes in market access.
3
Methodology
The central objective of this paper is to empirically relate bilateral migration rates
between the 27 Brazilian states to regional differences in market access. The derivation of the empirical specification of our migration equation used for this purpose is
presented in this section and follows Grogger and Hanson (2008) and Sorensen et al.
(2007).
We consider a log utility model as in Grogger and Hanson (2008). Every individual k coming from location i maximizes his indirect utility Vkij across all possible
destinations j. The individual location choice Mkij in a general utility differential
approach can then be written as:
Mkij = 1 if and only if Vkij = max (Vki1 , . . . , VkiR ),
= 0 otherwise.
The indirect utility Vkij can be decomposed as follows.
Vkij = Xij β + ξij + ekij
5
(18)
If expenditure for manufacturing goods is high, real wages might be high both because the
nominal wage is high and because the price index is low.
10
The utility of migrating to region j for an individual k from origin i is determined
by Xij , the characteristics of the location j. The product Xij β represents the utility
the individual receives from these characteristics, where β is a vector of marginal
utilities. The subscript i is included because some characteristics of location j can
vary across original locations, as for example bilateral distance.
The error term ξij represents unobserved location characteristics. Xij β and ξij
assign the same utility level to all individuals coming from i and going to j. In
order to still allow individuals from the same region to choose different locations, we
include an idiosyncratic error term that varies across both individuals and locations,
ekij . We assume that this error term follows an i.i.d. extreme value distribution.
Given that individuals choose the location that maximizes their utility, Equation
18 leads to:
P r(Vkij > Vkim )∀j 6= m
P r(ekij − ekim > Xim β − Xij β + ξim − ξij )∀j 6= m
(19)
McF (1974) shows that thanks to the i.i.d. extreme value distribution of the
individual error term, integrating out over the distribution of the logistic distribution
yields the following migration probabilities:
P r(Mkij = 1) =
exp(Xij β + ξij )
J
Σj=1 exp(Xij β + ξij )
(20)
Using the methodology proposed by Berry (1994), this probability of migration from
i to j can be interpreted as the share of individuals from i migrating to j. As in
Sorensen et al. (2007), we then write the share of migrants from i to j, sij , as
sij = P r(Mkij = 1) =
exp(Xij β + ξij )
J
Σj=1 exp(Xij β + ξij )
(21)
and the share of stayers of region i, sii , as
sii = P r(Mkii = 1) =
exp(Xii β + ξii )
J
Σj=1 exp(Xij β + ξij )
Dividing Equation 21 by Equation 22 and taking the log yields:
sij
exp(Xij β + ξij )
ln( ) = ln
= Xij β − Xii β + ξij − ξii
sii
exp(Xii β + ξii )
(22)
(23)
In section 5, we estimate this migration equation adding a time dimension t and
and sectoral dimension s. Further, we replace the vector X by our location-specific
and location-sector specific variables of interest. This yields:
ln mijst = ln
sijst
M Ajst
wjst
Xjt
= α1 + β1 ln
+ β2
+ β3 ln
+ ηt + ϕij + θs + vijst (24)
siist
M Aist
wist
Xit
where the dependent variable mijst is defined as the ratio of the number of
migrants in sector s in year t going from i to j over the number of stayers in sector
s. The ratios on M A and w correspond respectively to the state-sector specific
difference in international market access and wage levels. The construction of these
11
two variables is described in Section 4. ηt represents yearly fixed effects, θs represents
sector fixed effects and vijst is a i.i.d. bilateral error term. With the bilateral fixed
effects ϕij , we attempt to capture time invariant unobserved state characteristics
that are constant over time but might depend on the other states, i.e. distance.
The dummy for the couple ij (São Paulo - Amazonas) differs from the dummy for
the couple ji (Amazonas - São Paulo). These bilateral fixed effects take into account
time-invariant specificities concerning migration between two particular states. They
typically capture differences in climate, price indices or institutions (those which are
stable over our sample period) and thus proxy for moving costs between two specific
states (Greenwood, 1997).
Due to data limitations, we cannot add variables on the states’ amenities even
though we suspect them to play a significant role in shaping migration patterns
within Brazil. However, in our preferred specification we will add state-year fixed
affects for both state i and state j. These should capture time-varying differences
across states like price-level, amenities or institutional quality which could also be
important factors in the migration choice and risk to bias our results, when excluded.
As a consequence of controlling for bilateral and time fixed effects, we exploit the
panel dimension of our data. The coefficient on the ratio of market access indicates
how a change in the difference between the market access between two specific states
affects the evolution of migration between this pair of states.
A last important feature of our model is the independence of irrelevant alternatives (IIA). This implies that the probability of choosing one state relative to the
probability of choosing another is independent of the characteristics of a third state.
IIA arises from the assumption that the error terms are i.i.d. across alternative
destinations. IIA may be violated if two or more destinations are perceived as close
substitutes by potential migrants. Hausman and McFadden (1984) note that if IIA
is satisfied, then the estimated regression coefficients should be stable across all 27
destination sets. To test whether IIA is violated in our sample, we follow Grogger
and Hanson (2008) by re-estimating our model 27 times, each time dropping one of
the destinations. Coefficients of our market access and wage variables stay similar
across samples, suggesting that the IIA property is not violated here.
4
Data
In this section we will first present the data sources and how we calculate our
variables for migration rates, market access and region-specific wages.
4.1
Individual data
In this study, we use individual data for ten years from the period 1995 to 2003 (with
data missing for 1994 and 2000) collected and published by the Brazilian Institute
of Geography and statistics (IBGE).
Data come from the yearly household survey Pesquisa Nacional por Amostra de
Domicilios (PNAD), which covers between 310.000 and 390.000 individuals each
12
year.6 In 1994 the PNAD was not conducted because of a strike. 2000 was the year
of the population census which is conducted every ten years.
4.1.1
Migration rates
Our final sample will use migration flows constructed from the survey years 1995 to
2003.7 We define an individual as a migrant, when the answer given to the question
“In which Brazilian state did you live five years ago?” differs from the actual state
of residence.
Given our interest in bilateral migration within Brazil, individuals having lived
abroad in that period are excluded from our analysis. We further limit our sample
to individuals who declare to have a job and earn a wage above zero. Further, we
restrict are at least 20 years old at the time of the interview, but not older than
65. In line with migration literature, we restrict our sample for the calculation of
migration rates to information of household heads only.
In this study, we are interested in bilateral migration rates at the sectoral level.
Given that we don’t have information about the individual’s work five years ago,
we make the assumption that individuals already worked in the same sector as they
declare in the year of the survey. Bilateral migration rates are then defined as the
number of migrants from state i to j over the number of workers that stayed in
state i and declare working in sector s at the time of the interview. Sectoral market
access is computable only for 12 industries (manufacturing, mining and agriculture),
we hence limit our analysis to individuals who declare working in these industries.
The final samples of individuals that are used to construct bilateral migration rates
contains between 65,632 (1995) and 78,593 (2003) individuals for each round of the
PNAD.
To take into account heterogeneity of the impact of market access across different
types of workers, we calculate migration rates also according to skill levels. We
therefore distinguish between low and highly qualified individuals, where a person
is defined as qualified when she has more than eight or nine years of schooling.8
Column 1 in Table 1 reports the total number of migrants observed for each
year. On average 3.37% of all individuals in our final data set have moved at least
once within the last five year to another state, but migration has declined over the
sample period. Columns 2 and 3 report the numbers and percentages for low and
high qualified workers. Qualified individuals are throughout all years more mobile
than less qualified workers.9
Figure 1 displays differences in migrant shares between the two educational
groups for each state for the years 1993 and 2003. In 1993 migrants with higher education are dominant in at least one state in every Brazilian macro region.10 Whereas
6
The PNAD does not follow individuals, but interviews a different random (but representative)
sample of residents each year.
7
In the panel data set, we use 1995, 1996, 1997, 1998, 1999, 2001, 2002 and 2003.
8
We use the information on school years and on the highest degree the individual has obtained
or is enrolled in to identify the educational level of each individual.
9
Tables and figures in this section would be very similar also if we wouldn’t limit our analysis
to household heads or use a finer classification for education. However, they would be slightly
lower when including the nontradable sectors, remaining nevertheless clearly above the rates for
the low qualified workers.
10
The Brazilian states are grouped in five macro-regions (see also Table 3 and the map in Figure
13
Year
1995
1996
1997
1998
1999
2001
2002
2003
Mean
Table 1: Yearly migration rates
(1)
(2)
(3)
All
Low educated
Highly educated
Nb
%
Nb
%
Nb
%
2237
2022
1710
2292
2390
2584
2672
2522
3.41
3.21
3.50
3.42
3.45
3.37
3.40
3.21
1488
1329
1058
1419
1447
1532
1548
1396
3.07
2.93
3.11
3.02
3.01
3.00
3.05
2.81
3.37
3.00
749
693
652
873
943
1052
1124
1126
4.36
3.93
4.41
4.33
4.47
4.11
4.05
3.90
4.20
Source: Own calculations. Data are from the PNAD (1995 - 2003).
the Distrito Federal (53), where the capital Brasilia is located, Rio de Janeiro (33)
and the three states in the South (41 - 43), which are known for their good climate
and recent economic development, have stayed the preferred destinations of highly
educated workers, the North East and Center region (except the Distrito Federal),
have become the main immigration regions for low educated migrants.
These differences in the location choices suggest that the utility of migrating to
a specific state might vary across educational levels. They are also an indicator for
self-selection or for differences in location choice determinants across different types
of migrants.
When comparing mobility between sectors, we see that it varies substantially
across sectors as displayed in Table 2. Whereas in basic metals, textile and agriculture only 2.8% of the workers have moved to another state, this percentage is over
6% in coke, nuclear and petroleum and above 4% in food, beverages and tobacco,
pharmaceutical and the wood industry.
Despite the presence of a relatively high number of zero-flows between states,
the PNAD is considered as being representative of real migration rates and thus
adequate for studying migration patterns within Brazil (Fiess and Verner, 2002).
We will nevertheless address the problem of unobserved flows by running PoissonMaximum-Likelihood estimations including the zero-flows. For a detailed description of internal migration in Brazil see Fiess and Verner (2003).
A-1 in Appendix A.). This classification is based on the structural and economic development of
the different states, regrouping states with similar characteristics. The North is sparsely populated,
poor, and largely inaccessible. The Northeast is the poorest macro-region of Brazil with the lowest
life expectancy and wages, little access to mineral deposits or navigable rivers, and the highest
proportion of low educated persons. The Center-West combines a diverse set of characteristics,
mixing poor rural areas, dense forests, and the federal capital city of Brasilia, where income and
education levels are high. The Southeast and the South are the most economically developed
regions of Brazil. Education levels, income and life expectancy are all high in these regions,
and dense highway networks make it easy to get around. These regions offer high economic
opportunities and have a high population density.
14
Figure 1: Differences in migrant shares between high and low qualified workers
Source: Own calculations.
4.1.2
Regional wages
Our migration equation will also contain a sector and state specific wage variable.
Regional wage differentials are considered to be one of the key determinants of
migration. Also it is important to take into account differences in wage levels,
since, as described in Section 2, market access can simultaneously impact wages and
migration. The simplest way to include a wage variable in our specification would be
to take the average wage of each region. There are two reasons, not to do so. First,
given that an individual’s wage depend crucially on his educational attainment, it
would be better to allow the wage variable to vary across the educational levels.
Second, as highlighted in the literature, individuals currently living in state j are
not a random sample of the population, because similar individuals tend to move
to the same locations. This is reflected by the fact that immigrants in a specific
region often share common characteristics like gender, ethnic group or educational
level. Since these variables are also the main wage determinants, the wage level in
a region will correspond to the average characteristics of the people living there. As
a consequence, the average wage level is subject to a potential selection bias.
This problem is particularly important in the case of estimating migration as a
function of wages. The individual’s location choice is not a function of the average
wage level in each region, but depends on the wage she will gain in each region. Her
personal wage expectations are likely to differ substantially from the mean depending
on her personal characteristics. To circumvent this problem, Falaris (1987) proposes
15
Industry
Table 2: Migration rates by sectors
code Migration rates by sector
Average 1995
2003
Agriculture
Petroleum and gas
Mining
Food
Textiles
Wood
Paper & Printing
Chemical & Pharmaceutical
Plastic & non-metallic
Basic metals
Electrical & Electronics
Machinery
1
2
3
4
5
6
7
8
9
10
11
12
2,8
6,8
3,8
3,7
2,8
4,3
3,0
4,3
3,1
2,8
3,3
3,0
2,9
7,8
2,7
3,1
3,0
5,0
2,9
4,4
2,6
2,4
2,3
3,2
2,8
4,7
3,3
3,5
2,7
4,9
3,1
5,1
2,8
2,2
3,3
3,5
Population by sector
1995
2003
11788
64
329
1895
1739
695
516
681
996
2114
310
1028
12221
85
364
1710
2195
769
548
861
968
2278
338
1179
to use predicted wages that have been estimated by a Mincerian wage equation using
an additional correction term, that accounts for self-selected migration.
To obtain a wage variable that corrects for self-selected migration, we follow
closely De Vreyer et al. (2009). These authors generalize the approach of selfselection correction developed by Dahl (2002), who corrects the estimates of individual wages by adding the individual’s migration probabilities into the state-specific
wage equation.
In a first step, we thus need to determine migration probabilities of all individuals
in our data set. For this, we estimate a multinomial logit model based on the
individual’s location choice.11
This allows us to obtain the probabilities of an individual to migrate to each of
the Brazilian states as a function of his personal characteristics. A polynomial of
these migration probabilities are then added as additional regressors in the wage
equations to get consistent estimates of the coefficients.12
The obtained estimates are then used to predict wages for each individual in
each of the 27 states.
We are interested in comparing the wage an individual gains in the sector in her
actual state of residence with the wage she could gain in the same sector in any
other of the Brazilian states.
11
For the sake of brevity, regression results used for the construction of the wage variable
are not reported but are available on request. Traditional selection bias corrections (like the
conditional logit model) are not very well suited to cases where individual migration decisions
imply numerous potential destinations. We thus run a multinomial logit separately for each year,
predicting individual migration probabilities for each state based on the individual’s educational
level, age group, family status, state of origin and a dummy for migrants.
12
The wage equation includes the individual’s education, gender, age, ethnic group and its
location (rural vs urban). As in Dahl (2002), we include as correction terms the individual’s
probability of living in the state of destination and the probability of staying in the original
state. We try different specifications of the wage equation including more migration probabilities
and different number of polynomials, but find only small differences in predicted wages. Wage
equations are run separately for each year.
16
In section 5, we estimate bilateral migration rates at the sectoral level. Therefore,
we need use also sector-state specific wage variables. The aggregation of individual
wages to sector-state average wages is done in the following way: The wage variable
for the state of origin i, wis , will be the average of all workers in sector s who
actually lived in i five years ago (more precisely, the mean of their predicted wages).
In contrast, the wage variable for a given destination j, wjs , varies according to the
state of origin of the migrants. For each destination j, we compute the wage variable
wjs according to the origin of the residents working currently in sector s. The wage
variable of São Paulo thus differs between individuals coming from Amazonas and
those coming from Acre. For the former we take the mean of the predicted wages for
São Paulo of all the workers declaring Amazonas as the origin, and for the later we
take the mean of the predicted wages for São Paulo for the individuals having lived
in Acre. This method allows us to use the mean of predicted wages of the same
individuals for computing the regional wage at the origin and at the destination,
what reduces again any potential selection bias.
We compute the wage variable once using the predicted wages corrected by selfselection bias and once using predicted wages that do not take into account the
additional correction term. Both variables are highly correlated and in the empirical
analysis, results obtained with the corrected and with the uncorrected wages are
mostly the same, so in the interest of space, we will report and discuss only results
with corrected wages.13
In section 5.3, we distinguish migration rates and wages in addition according to
the educational level of migrants. The wage variables are obtained in the same way
as at the sectoral level, but take different values for high and low educated workers.
By introducing the state’s wage level in our migration equation, we also control
for an impact of market access on wages. The market access impact we observe in
our regression is the effect on migration beyond the effect on wages.
4.2
Market Access Calculation
Our main variable of interest is the state’s market access for every specific sector,
as defined in Equation 7 in Section 2. To obtain a sound estimate of market access,
we follow the methodology pioneered by Redding and Venables (2004).
First, we estimate the following gravity trade equation14
φ µE Gσ−1
EXij == ni pij qj = si φij mj = ni p1−σ
| {zi } ij | j{z j }
F Xi
(25)
F Mj
where EXij are bilateral exports from region i to region j. i and j can be either
a Brazilian state or one of the other 170 countries included in the trade data set.
13
Note that we are using nominal wages here, which are deflated for all years using the Brazilian
consumer price index. The pertinent variables for migration should be real wages that are deflated
by region-specific price indices. Unfortunately, information on state-specific price indices covers
only main Brazilian cities in a limited number of states. By including year-state fixed effects in
our final regression, we are aiming to control for differences in price indices between states and
over time.
14
Detailed derivation of the trade gravity equation, the market access construction and
additional estimations can be found in Paillacar (2006) and Fally et al. (2008), available at
http://team.univ-paris1.fr/teamperso/paillacar/.
17
Under the assumption of homogeneous firms, total exports from i to j can be written
as a function of the quantity qij sold at the price pij on market j by the number of
firms in i, ni .
The first term on the right hand side of the equation, ni p1−σ
, is considered to
i
be the supply capacity of region i. It is defined as the number of firms active in i
times the f.o.b. price. Supply capacity and market capacity are country-specific for
country i and country j and can be captured by exporter and importer fixed effects,
F Xi and F Mj . The freeness of trade variable, φij , can be proxied by different
variables that enhance or deter trade such as bilateral distance dij and contiguity
Cij .
For the calculation of a sector-region specific market access variable, we estimate
equation 25 separately for every year for each of the 12 sectors.15
Taking the logs of equation 25 and adding the sectoral dimension s and the time
dimension t, our estimated specification of the trade equation becomes
ln EXijst = F Xist +F Mjst +δst ln dij +λ1st Cij +λ3st RT Aij +λ4st W T Oij +uijst (26)
The freeness of trade, φij , is captured by the bilateral distance dij , the contiguity
Cij , the presence of a free trade agreement between the two trading partners RT Aij
and whether the two are member of WTO (or its predecessor GATT), W T Oij . uijst
is a random bilateral error term.
Once we have the estimates of bilateral transport costs φij and market capacity
of each foreign trading partner mj , we can put these into equation 7 in Section 2
and obtain a foreign market access for every Brazilian state.
Note that since we estimate the gravity equation separately for each sector and
year, all coefficients and fixed effects are allowed to vary across sectors and years,
allowing us to build a state-sector specific market access indicator for every year of
our sample period.
Our migration panel data set covers the years 1995 to 2003. We calculate market
access for the years 1991 to 2002. In order to explain for example the migration
between 1990 and 1995, we then take the mean of the market access variable for the
years 1991 to 1994.
This methodology is rarely applied in regional studies because of data limitations:
bilateral trade flows are rarely available at the intranational level, particularly for
developing countries. Brazil is a fortunate exception since it provides information on
international trade flows for each of the 27 states and also the sectoral composition
of the flows.16
Table 3 regroups all information into one indictor by state, which is the average
rank of all sector-specific market access indicator the state obtained for each in the
eight years of our analysis. There are important differences in the ranking across
sectors and states. Overall the state which has in most sectors and years the lowest
15
Data is available at ISIC Rev 3 at 2-digits. The classification used in the official statistics
is the Brazilian nomenclature CNAE 3.1 which is fully equivalent to ISIC Rev 3 at the level of
aggregation that we are considering (See Table A-1 in Appendix A for the full list). Average
values of the estimated coefficient across industries are available at Fally et al., 2008, Table 2,
second column
16
Since intranational trade flows are available only for 1999, the market access variable will
be estimated based on international trade flows only in order to be consistent over years. For a
detailed description of the data and its sources, please refer to Appendix B.
18
Rank of foreign
market access
11
12
13
14
15
16
17
Rondônia
Acre
Amazonas
Roraima
Pará
Amapá
Tocantins
7,7
6,3
3,7
2,1
6,8
4,7
14,2
Northeast
21
22
23
24
25
26
27
28
29
Maranhao
Piauı́
Ceará
Rio Grande do Norte
Paraı́ba
Pernambuco
Alagoas
Sergipe
Bahia
9,1
12,0
11,3
13,2
15,5
16,8
18,8
18,6
19,1
Southeast
31
32
33
35
Minas Gerais
Espı́rito Santo
Rio de Janeiro
São Paulo
19,6
21,5
19,5
17,6
South
Code
41
42
43
Paraná
Santa Catarina
Rio Grande do Sul
9,1
10,4
4,8
50
51
52
53
Mato Grosso do Sul
Mato Grosso
Goiás
Distrito Federal
11,4
12,3
15,7
17,0
North
State
Center-West
Table 3: International market access by state (mean over sectors and years).
Source: own calculations.
market access is Espı́rito Santo in the Southeast, which has an average rank of 21,5.
Roraima, the most northern state has with 2,1 the highest average ranking over the
sample period.
To test the robustness of our measure, we also compute different market access
indicators. First, estimate equation 26 using different specifications for φij . Second, we obtain a market access indicator that includes also a domestic dimension.
Although a complete indicator of market access, including the import capacity of
all Brazilian states would be interesting to study, as described in the introduction,
employing only the international part of market access has the important advantage
19
of reducing endogeneity. However, theory predicts that all markets should matter for the migration choice and thus national and international demand should be
accounted for. We therefore proxy the spatial distribution of national demand by
including distance weighted importer fixed effects obtained for the state’s in our
regressions. Even though these effects capture only the demand for imports, they
can be seen as a good proxy for local demand. A further drawback of this method
is, that not every state is importing in every year goods in all sectors. In order to
have market access defined by the same partners in every year and every sectors, we
include only the importer fixed effect for the eight states for which we can estimate
the sector-specific market capacity in all cases. Nonetheless, since these eight states
represent well the overall distribution of economic activity within Brazil, we believe
that this measure is a good proxy for the evolution of the total market access. In
addition, in the regression analysis, we will also capture local demand by the state’s
income per capita and population, which allows also to control for a possible positive
correlation between international and local market access.
As a last test for the robustness of our market access indicator, we address the
concern of zero-value flows. Both the survey data from which we obtain migration
rates and the trade data we use to compute our market access variable are characterized by a high number of zero-value flows. In this context, OLS regressions on
our gravity equations can be criticized because they lead to biased estimates in the
presence of many zero value flows. A high number of zero value trade flows is likely
to result in biased estimates of the parameters used to compute market access. To
treat zero-flows in a gravity equation researchers often recur to non-linear estimations (e.g. Gamma or Poisson regressions) that allow to deal with some specific
problems of trade data.
A definitive method on how the problem of zero-flows should be treated has however not yet emerged (Silva and Tenreyro (2006); Martinez-Zarzoso et al. (2007);
Martin and Pham (2008); Helpman et al. (2008); Silva and Tenreyro (2009)). The
question is even more difficult to answer when non-linear estimations are combined
with location fixed effects (Buch et al., 2006). When the proportion of zeros becomes too important, iterative methods for Poisson or Gamma estimations often do
not converge. This problem appears to be particularly acute, when we run trade
regressions at the sectoral level. The proportion of zeros can easily surpass 80%,
and iterative methods for Poisson or Gamma estimations often do not converge.
We therefore follow Helpman, Melitz and Rubinstein (2007) and control for zeros
by including a first step and estimate first a selection equation, where our selection
variable is the “doing business” indicator obtained from the World bank.
5
Results - Market Access and migration response
In this section, we estimate the migration equation, equation 24 in Section 3, using
standard panel data estimation techniques and Poisson maximum likelihood regressions. We look at regional differences in sectoral market access to evaluate the role
of sector-specific market access and heterogeneity in the reaction to market access
across manufacturing sectors.
20
5.1
Sector-specific market access and migration rates
In this section, we test empirically whether an increase in the gap of market access
between two states affect the bilateral migration rates between these two states. Our
key variable of control is the regional difference in wages, which are also calculated
at the sectoral level. Since market access also has a positive impact of wages, we can
interpret our coefficient on market access as the impact this variable has beyond its
positive impact on the local wage level.
Our key independent variable is the ratio of the destination state’s foreign market
access for sector s over the foreign market access for sector s of the state of origin
(M Ajs /M Ais ). We expect market access of the origin, M Ais , to have a negative
impact on the bilateral migration rate: the higher M Ai , the higher the demand
for work. Thus, the individual can attain already a high utility in his home region
and is thus less likely to look for (and find) a better job in another state. For
the destination’s market access (M Aj ) we expect a positive impact, since a higher
market access reflects a higher wage level and a higher number of available jobs as
a response to the increase in foreign demand. Therefore, a higher market access
should attract more people in search for a job in export-oriented industries. The
coefficient of the market access ratio is thus expected to be positive.
To take into account that market access and regional wages are estimated in
a prior regression, displayed standard errors, which are clustered at the origindestination-couple-level, are bootstrapped.
We start by estimating a general model with all workers and pooling all industries. The first three columns in Table 4 present standard panel regressions including
dyadic fixed effects (F Eij ) and time and industry fixed effects. In column 1, we first
want to know, how our specification performs in explaining bilateral migration rates
without our variable of interest.
In this specification we consider standard migration determinants with sectorregion specific wages as only sector-specific variable (see section 4.1.2). Our second
key control variable is the difference in unemployment rates. We consider that
the primary channel by which high market access attracts migrants are the good
employment opportunities, including not only the type of job but also the high
number of available jobs. This last component is typically also reflected by the local
unemployment rate. Whereas our market access indicator should be a determinant
of contracted as well as speculative migration, the unemployment rate is mostly
important for the speculative type of migration.
Since official unemployment data at the state level is not available, we compute
a local unemployment rate with the information of the survey data. 17
17
Official regional unemployment rates are only available for a few big cities on the coast.
21
22
0.792
R2
0.794
3089
yes
-0.039
(0.075)
-0.225
(0.209)
0.818
3089
yes
15143
yes
-0.131
(0.210)
-0.744
(0.563)
3.011
(2.057)
-0.522b
(0.233)
0.051
(0.032)
1.160a
(0.324)
FMA
(4)
Clustered standard errors in parentheses
All regressions contain industry and dyadic fixed effects.
c
p<0.1, b p<0.05, a p<0.01
3089
Observations
-0.039
(0.076)
deathjt /deathit
yes
-0.244
(0.210)
gdpjt /gdpit
F Eit &F Ejt
F Et
0.857
(0.940)
popjt /popit
0.646
(0.925)
-0.190c
(0.097)
-0.173c
(0.098)
ujt /uit
0.205a
(0.032)
0.586c
(0.354)
0.612b
(0.310)
0.169a
(0.027)
FMA
OLS
FMA
0.180a
(0.027)
(3)
(2)
wjts /wits
M Ajts /M Aits
(1)
0.072b
(0.035)
15143
yes
(6)
15143
yes
-0.135
(0.210)
-0.715
(0.565)
3.073
(2.061)
-0.538b
(0.235)
0.052
(0.033)
0.845a
(0.228)
Poisson
Heckman
0.931a
(0.319)
FMA
(5)
15143
yes
0.073b
(0.036)
0.723a
(0.233)
Heckman
(7)
0.826
3089
yes
0.176a
(0.028)
0.442a
(0.057)
OLS
TMA
(8)
Table 4: : Bilateral Migration - benchmark ijk
15143
0.071b
(0.034)
0.252a
(0.077)
Poisson
TMA
(9)
Finally, we introduce also population and GDP per capita.18 Both variables
can be seen as a proxy for local economic activity and are strongly correlated with
our market access variable (see Table ??). It is thus possible, that the significant
coefficient of our variable of interest captures, at least partially, the impact of the
evolution of population or of the per capita income. A last variable is the ratio of
the homicide rate in the state of destination and state of origin, which we consider
as a proxy of crime and security of a state.
Coefficients of sector-state specific wages and state-specific unemployment rates
are significant and have the expected signs. Somewhat more surprisingly, we cannot
confirm any positive impact of population or GDP differentials on the migration
decision. This is due to the fact that these variables are defined at state level, while
our dependent variable is defined at the industrial-state level.
In the second column we add our variable of interest, the gap of foreign market
access. In the third column, we include year-state fixed effects for the origin and the
destination. This allows to control for time and state varying variables like the price
index, amenities and other variables that are omitted from Column 2. In Column 3,
we then find that even with this full battery of fixed effects, the coefficient of market
access is still significant at the 10% level. The next four columns confirm the results
when the Poisson Pseudo-Maximum Likelihood (PPML) is applied to deal with zeromigration flows. Columns 4 and 5 repeat the specifications of Column 2 and 3 and
obtain qualitative similar results. The coefficient of the foreign market access gap
is very high whereas we see a lower and less significant impact of the wage gap than
in the panel regressions. Columns 6 and 7 confirm results using the market access
variable obtained from estimating the Heckman selection model, which controls for
the presence of zero trade flows that could bias our estimates for M A.
In the last two columns, we use as market access variable an indicator that
includes also a proxy for domestic demand. Our key independent variable will
thus be total market access, T M A. Also these estimations confirm a positive and
significant effect of the market access differentials in the shaping of internal migration
flows. Nevertheless, coefficients are much lower than comparable specifications with
the foreign market access gap.
A preliminary conclusion on this first section thus suggests that domestic and
in particular international market access can be considered as a good indicator for
export-oriented economic opportunities and as a determinant of bilateral migration
rates.
5.2
Industries with comparative advantage
Since we have migration rates at the sectoral level, we can split our sample into
groups. Whether an industry is in decline or flourishing following the opening to
international trade, is likely to the role market access is plating in the migration
decision of workers, depending on which type of industry he is working in. Muendler
et al. (2010) study job flows in Brazil and confirmed that tariff reductions affected
hirings negatively and job separations positively. More interestingly, they found
that sectors with an international comparative advantage tend to fire more and
18
Data on population and GDP for the Brazilian states are from the IPEA (Instituto de Pesquisa
Econômica Aplicada).
23
hire fewer workers than the average sector, which goes in the opposite direction of
ideas proposed by traditional trade models. In our case, and assuming that worker
mobility across sectors is low (as the evidence by Muendler et al. 2010 suggests)
we could expect that comparative advantage sectors reinforce the potential of a
region to attract workers. In principle, we could also postulate that the reaction
in declining sectors should be similar (or even stronger), because workers in those
sectors would like to find shelter from competition and hence, privilege high market
access locations. Nevertheless, it is also possible that workers in declining sectors
face some credit constraint to move.
To define whether an industry is enjoying a comparative advantage on the international market or not, we employ the measure of comparative advantage for
Brazilian industries proposed by Muendler (2007). He computed the Balassa indicator of revealed comparative advantage for every industry at the ISIC 2 level. The
Balassa indicator is defined as the ratio of the export share of a specific industry
in Brazil and the average export share of the same industry at world level. A sector with a ratio higher than 1 is said to have a Revealed Comparative Advantage
(RCA). As this differentiation may be considered as too strict, it is safer to use the
distribution of the RCA indicator. Muendler reports the distribution in quintiles
and comments that industries are stable across quintiles during the period 1990-97.
Table A-1 lists for every industry the value of this indicator as well as the quintile
of the distribution in which the sector is located. We classify the different industries
according to their comparative advantage into two groups. Industries in the comparative advantage group are those whose Balassa indicator lies in the fifth or fourth
quintile of the distribution. The remaining industries are considered as not having a
comparative advantage. As a robustness check, we split this group and run separate
regressions for industries in the third quintile and industries in the remaining two
lowest quintiles. We did not find any relevant difference with respect to the results
in the tables displayed in the paper.
Results reported in Column 1 to 4 in Table 5 confirm that workers in the sectors
at high comparative advantage are driving the results in the case of international
market access. The coefficient of international market access in those industries
ranges between .8 and 1, depending on the chosen econometric specification. This
means that workers in declining industries and located in unfavorable regions are
not considering migration as a strategy to cope with the decadence of the industry.19
By contrast, workers in winning industries are moving to the better located regions,
to fully take advantage of the positive economic prospects in their industry.
Although a close relationship between migration and F M A in sectors with comparative advantage can be expected, domestic demand of the specific industry should
also matter. However, when including domestic demand, the impact of the market
access indicator on the migration decision is expected to be less related to the
comparative advantage related to exports. Columns 5 to 8 in Table 5 repeat the
estimations displayed in Column 1 to 4 but using our measure of total market access.
19
Other alternatives to cope with a negative impact of trade openness are (1) sectoral mobility
(within tradable sectors) and (2) entering into non-tradable sectors (in particular informality or
services). Muendler et al. (2004) have shown that the first channel is weak in Brazil, especially
among flows from losing sectors towards wining sectors. By contrast, the same authors found
evidence of the second channel.
24
This shows that there is no significant difference between both groups of industries
when taking into account also the demand pattern within the country. Hence, we
see that also workers in the sectors that do not enjoy a comparative advantage in
international trade react to changes in the spatial distribution of demand. When a
state is facing an increasing demand for its goods on the national market (even if
the sector is not successful exporter at national level) then firms will also need to
recruit more labor and thus it is the domestic demand that drives the positive result
for total market access here.
25
26
2140
0.817
Observations
R2
9093
yes
0.157
(0.125)
1.066a
(0.410)
949
0.920
yes
0.102b
(0.048)
0.275
(0.479)
6050
yes
0.081b
(0.041)
0.626
(0.580)
2140
0.817
yes
0.424a
(0.106)
0.260a
(0.076)
9093
yes
0.122
(0.126)
0.366a
(0.100)
949
0.923
yes
0.091c
(0.047)
0.371a
(0.121)
6050
yes
0.079b
(0.039)
0.336
(0.208)
(7)
(8)
No advantage
Bootstrapped standard errors in parentheses (1000 replications), clustered at the origindestination level. All regressions contain dyadic and industry fixed effects.a , b and c represent
respectively statistical significance at the 1%, 5% and 10% levels.
yes
0.460a
(0.105)
0.854c
(0.477)
F Eit &F Ejt
wjts /wits
T M Ajts /T M Aits
F M Ajts /F M Aits
Table 5: The role of comparative advantage
(1)
(2)
(3)
(4)
(5)
(6)
Comp advantage
No advantage
Comp advantage
Overall, these findings contribute to a better understanding of the NEG forces at
work in the spatial economy. Though we observe the agglomeration effect described
in NEG theory (higher demand attracts new workers and leads to bigger agglomerations), this mechanism is not affecting all individuals in the same way. Workers
will choose their destination depending on their sector and the spatial structure of
their industry’s market access. As a consequence, the biggest agglomeration will not
necessarily attract all workers. It is more likely that individuals will go to different
locations, depending on the region’s sectoral composition. Once a region has acquired an advantage in a certain industry (due to comparative advantages or other)
and its market access increases for this industry, specialization will be facilitated,
because the high sectoral market access will attract corresponding workers. We
hence observe an adjustment by the quantity mechanism at the sectoral level.
5.3
High versus low educated migrants
In this section, we split our sample according to the two levels of education. This allows us to see whether low or highly educated migrants are more sensitive to changes
in market access. The migration literature stresses the potential heterogeneity in
migration behavior of skilled versus unskilled workers. Several studies note that
skilled workers may be more sensible to other regional characteristics like amenities
(see for e.g. Tabuchi and Tisse 2002) and this could indeed act as an opposing force
to market access, especially if the agglomeration induced by a high market access
provokes congestion costs. While workers are attracted for economic reasons to a
certain region, non-economic reasons could retain them in the origin region or redirected them to other regions with better amenities.20 This mechanism can reduce
the potential for industrial agglomeration allowing regional wage disparities to persist. Given that we can’t control exhaustively for amenities (beyond the variable of
the state’s homicides rate which we consider as a proxy of crime and security of a
state), this is a particularly important aspect that justifies a separated analysis by
skill level.21
Differentiating among skill levels is also important for reasons linked to international trade. According to Redding and Schott (2003), who predict a higher wage
premium due to good market access for skilled workers, we could expect that highly
educated workers have a stronger incentive to go to regions with good access to foreign markets. Nevertheless, this argument may not be valid in the case of un-skilled
labor abundant developing countries. As Brazil is a country relatively abundant in
unskilled labor (see Muriel and Terra, 2009, for evidence on this), we expect that
relative returns to this factor increase in the case of trade liberalization. Gonzaga
et al. (2006) confirm this prediction by showing that wages in Brazil for unskilled
workers have increased during the period 1988-1995. However, these authors stay
20
For example, Levy and Wadycki (1974) have shown that in Venezuela educated individuals
tend to value amenities much more than low qualified and Schwartz (1973) argues that the negative
impact of distance on migration flows diminishes with educational attainment.
21
The literature also notes that skilled workers are more mobile because they face less credit
constraints to finance the migration, they suffer from less asymmetric information on economic
opportunities and they are less risk averse. This differentiated propensity to migrate among skill
levels is also found in the case of Brazil. Nevertheless, the interest of this paper is to analyze the
determinants of the bilateral migration response rather than the propensity to migrate or not.
27
with their analysis at the national level. At the subnational level, using microdata
for 1999, Fally et al (2010) have found that the international component of market
access raises wages of unskilled workers relatively more than wages of skilled workers. Further, their study points out that this differentiated effect does not apply to
total market access, ruling out a role of domestic demand, but is driven exclusively
by the state’s and sector’s access to foreign markets. Fally et al.’s findings thus
suggest that international trade affects factor returns (in particular wages) with a
combined effect of locational advantages (market access premium) and a traditional
Stolper-Samuelson mechanism (higher payment to the relatively abundant factor).
Results for international market access are reported in Table 6. The first three
columns report results for low qualified workers with no more than nine years of
schooling, the last three columns contain results for workers with more than nine
years of schooling. Column 1 and 4 show panel regressions employing origin-year
and destination-year FE in addition to dyadic, time and industry fixed effects. This
allows to obtain coefficients for state-industry specific variables (wages and market
access). The remaining columns report Poisson estimates. Only for migration rates
of unskilled workers the coefficient of market access is significant at conventional
levels. By contrast, results for total market access presented in Table 7 suggest a
more similar effect on migration rates for both skill levels. In the standard panel
specifications the coefficient on the market access gap is significant for both types of
workers. However, in our preferred specification (Column 2 and 4), the coefficient
for skilled workers is again not significant, confirming overall the results from the
previous table, that market access is a driving factor in the migration decision
mainly for un-skilled workers. The absence of a differentiated effect by skill levels
is consistent with the argument that there is no skill premium associated to market
access when domestic demand is included.
The differences in migration patterns between educational levels can thus be
explained in part by their different sensitivity to market access. Economic opportunities associated to international trade are most important for the location choice
of low educated individuals. The findings of no impact of changes in market access
is in line with the migration literature cited above, highlighting that with a higher
educational achievement non-pecuniary benefits are crucial for the location choice.
Also, since Brazil is exporting mainly goods that are intensive in unskilled-labor,
an increase in exports will lead to an increase in demand for low educated workers.
Maia (2001) studies the impact of trade and technology on employment in Brazil
and finds that both were destroying more unskilled than skilled job. Low-educated
workers are thus more likely to be touched by the loss of their previous job and thus
are obliged to move to another state to find a new employment.
The importance of the international and national component confirm that including only GDP per capita or other state characteristics that represent only the
local economic activity does not reflect the migration decision of those who come
because of the location’s opportunities related to its relative location in the country
or in the world economy.
28
Table 6: Bilateral Migration by sector and skill level - Foreign Market Access
(1)
(2)
(3)
(4)
(5)
(6)
Low skilled workers
OLS
Poisson Poisson
Highly skilled workers
OLS
Poisson Poisson
F M Ajts /F M Aits
0.803c
(0.456)
1.345a
(0.492)
1.140b
(0.509)
0.052
(0.277)
0.551
(0.414)
0.487
(1.066)
wjts /wits
0.517a
(0.110)
0.196b
(0.096)
0.186
(0.129)
0.123a
(0.036)
0.062c
(0.033)
0.051
(0.075)
ujt /uit
1.620
(2.843)
-6.805c
(3.896)
popjt /popit
0.147
(0.641)
-0.710
(0.941)
gdpjt /gdpit
-0.340
(0.256)
-0.045
(0.357)
deathjt /deathit
-0.398c
(0.239)
-0.246
(0.268)
F Eit &F Ejt
F Et
Observations
R2
yes
no
no
yes
yes
no
yes
no
no
yes
yes
no
2282
0.858
11477
11477
1058
0.905
7961
7961
Clustered standard errors in parentheses
All regressions contain industry fixed effects and dyadic fixed effects.
c
p<0.1, b p<0.05, a p<0.01
Table 7: Bilateral Migration by sector and skill level - Total Market Access
(1)
(2)
(3)
(4)
Low skilled workers
OLS
Poisson
Highly skilled workers
OLS
Poisson
T M Ajts /T M Aits
0.388a
(0.067)
0.242c
(0.143)
0.239a
(0.082)
0.344
(0338)
wjts /wits
0.452a
(0.102)
0.194
(0.132)
0.120a
(0.035)
0.051
(0.075)
F Eit + F Ejt
yes
yes
yes
yes
Observations
R2
2282
0.861
10742
1020
0.909
7629
All regressions contain industry, time and dyadic fixed effects.
Standard errors in parentheses. c p<0.1, b p<0.05, a p<0.01
29
5.4
High versus low educated migrants and comparative advantage
Results so far have shown that international market access has effects the migration pattern of low skilled workers and for workers that have a job in an industry
that is considered to enjoy a comparative advantage on the world market. These
aspects are consistent with the effects that could be expected from Brazil’s recent
trade liberalization. By contrast, total market access, which includes a proxy for
domestic demand, seems to affect all workers in similar ways and independently of
the evolution of the comparative advantage of the industries.
In this section we explore jointly the heterogeneity across sectors and skill levels.
Columns 1 and 3 of Table 8 show the results of testing the impact of international
market access for low skilled workers in comparative advantage industries. Columns
2 and 4 give the results for low educated workers in comparative advantage industries. Columns 5 to 8 report results for highly skilled workers. Table 9 repeats the
same regressions using total market access instead of foreign market access.
Several conclusions can be drawn from these results. First, there is no support
for the hypothesis of an impact of market access (international or total) for skilled
workers. However, this result must be taken cautiously, as the number of observations for this group is low, and the analysis in the previous sections suggests that the
version of market access including the domestic demand influenced skilled workers
if all industries are pooled.
Second, and more interesting, we confirm that the change in the difference in
international market access bilateral gaps contributes significantly to the explanation
of migration patterns of unskilled workers in the winning industries.
30
31
(8)
1747
0.854
Observations
R2
535
0.952
yes
7651
yes
0.274
(0.173)
0.999b
(0.471)
3826
yes
0.190
(0.165)
0.505
(0.949)
616
0.933
yes
0.139
(0.100)
-0.201
(0.456)
442
0.962
yes
0.109
(0.080)
0.343
(0.592)
Standard errors in parentheses. All regressions contain industry and dyadic fixed effects.c p<0.1,
yes
0.051
(0.306)
0.756a
(0.177)
wjts /wits
F Eit + F Ejt
0.546
(1.664)
1.110c
(0.611)
F M Ajts /F M Aits
b
p<0.05,
4404
yes
-0.043
(0.137)
-0.229
(0.488)
a
p<0.01
3557
yes
0.054
(0.046)
-0.113
(0.603)
Low skilled workers
Highly skilled workers
Comp adv No adv Comp adv No adv Comp adv No adv Comp adv No adv
OLS
Poisson
OLS
Poisson
Table 8: Bilateral Migration by sector and skill level - Comparative advantage - FMA
(1)
(2)
(3)
(4)
(5)
(6)
(7)
32
(8)
1747
0.853
Observations
R2
535
0.955
yes
7651
yes
0.246
(0.171)
0.306a
(0.114)
3826
yes
0.215
(0.163)
-0.094
(0.270)
616
0.934
yes
0.144
(0.098)
0.210
(0.133)
c
p<0.1,
442
0.963
yes
0.098
(0.077)
0.324
(0.268)
Standard errors in parentheses. All regressions contain industry and time fixed effects.
yes
0.016
(0.286)
0.743a
(0.187)
wjts /wits
F Eit + F Ejt
0.507
(0.358)
0.238a
(0.081)
T M Ajts /T M Aits
b
p<0.05,
4404
yes
a
-0.053
(0.134)
0.222
(0.194)
p<0.01
3557
yes
0.054
(0.046)
0.022
(0.256)
Low skilled workers
Highly skilled workers
Comp adv No adv Comp adv No adv Comp adv No adv Comp adv No adv
OLS
Poisson
OLS
Poisson
Table 9: Bilateral Migration by sector and skill level - Comparative advantage - TMA
(1)
(2)
(3)
(4)
(5)
(6)
(7)
6
Conclusion
This study analyzes the impact of trade on internal migration decisions in a New
Economic Geography framework by looking at the relationship between market access and bilateral migration rates. We regroup migrants into 12 sectors and two
educational levels and show how migration patterns and reaction to changes in
sector-specific market access vary across different types of workers.
We see that access to markets plays an important role in the migration decision
beyond the effect that market access has on wages which are themselves important
determinants of migration flows. States with low market access push residents to
migrate to states with higher market access, where higher labor demand offers more
jobs and higher wages. Low educated workers seem to be more sensible to changes
in market access than highly educated workers.
Overall, we see that sector-state specific market access plays a more important
role in the migration decision for workers in all sectors, but that foreign demand
is important only for workers in sectors with a high comparative advantage on the
international market. Our robustness checks, include count data methods which
confirm a positive and significant impact of the regional gap in market access in
shaping bilateral migration patterns.
The fact that migration patterns are driven by industrial specialization suggests
that implications of NEG theory (i.e. regional advantages generated by the region’s
position in the spatial economy) are better understood in combination with comparative advantage and sector-specific inputs (e.g. human capital specificity).
33
References
(1974). Frontiers of Econometrics, Chapter Conditional Logit Analysis of Qualitative Choice Behavior, pp. 105–142. Academic Press.
Aguayo-Tellez, E., M.-A. Muendler, and J. P. Poole (2008). Globalization and formal
sector migration in Brazil. WIDER Research Paper 22, UNU-WIDER.
Berry, S. T. (1994). Estimating discrete-choice models of product differentiation.
RAND Journal of Economics 25 (2), 242–262.
Breinlich, H. (2006). The spatial income structure in the european union - what
role for economic geography? Journal of Economic Geography 6:5, 593–617.
Buch, C. M., J. Kleinert, and F. Toubal (2006). Where enterprises lead, people
follow? Links between migration and FDI in Germany. European Economic
Review 50 (8), 2017–2036.
Combes, P.-P., G. Duranton, and H. Overman (2005). Agglomeration and the adjustment of the spatial economy. Papers in Regional Science 84 (3), 311–349.
Crozet, M. (2004). Do migrants follow market potentials? An estimation of a New
Economic Geography model. Journal of Economic Geography 4 (4), 439–458.
Dahl, G. (2002). Mobility and the return to education: Testing a Roy model with
multiple markets. Econometrica 70 (6), 2367–2420.
De Vreyer, P., F. Gubert, and F. Roubaud (2009). Migration, Self-selection and Returns to Education in the WAEMU. Journal of African Economies Forthcoming.
Falaris, E. M. (1987). A nested logit migration model with selectivity. International
Economic Review 28 (2), 429–43.
Fally, T., R. Paillacar, and C. Terra (2008). Economic geography and wages in Brazil:
Evidence from micro-data. Technical report. Available at http://team.univparis1.fr/teamperso/paillacar/.
Gould, D. M. (1994). Immigrant links to the home country: Empirical implications
for U.S. bilateral trade flows. The Review of Economics and Statistics 76 (2),
302–16.
Greenwood, M. (1997). Internal migration in developed countries. In M. Rosenzweig
and O. Stark (Eds.), Handbook of Population and Family Economics, Volume 1,
Chapter 12, pp. 647–720. Amsterdam: Elsevier.
Grogger, J. and G. H. Hanson (2008). Income maximization and the selection and
sorting of international migrants. NBER Working Paper 13821.
Hanson, G. (1998). Regional adjustment to trade liberalization. Regional Science
and Urban Economics 28 (4), 419–444.
Harris, J. and M. Todaro (1970). Migration, unemployment and development: A
two-sector analysis. American Economic Review 60 (1), 126–142.
34
Hausman, J. and D. McFadden (1984). Specification tests for the multinomial logit
model. Econometrica 52 (5), 1219–1240.
Head, K. and T. Mayer (2004). The Empirics of Agglomeration and Trade. In
V. Henderson and J. F. Thisse (Eds.), Handbook of Regional and Urban Economics, Volume 4, pp. 2609–2669. Amsterdam: Elsevier.
Head, K. and T. Mayer (2006). Regional wage and employment responses to market
potential in the EU. Regional Science and Urban Economics 36 (5), 573–594.
Head, K. and J. Ries (1998). Immigration and trade creation: Econometric evidence
from Canada. Canadian Journal of Economics 31 (1), 47–62.
Helpman, E. (1998). The size of regions. In S. E. Z. I. Pines, D. (Ed.), Topics in
Public Economics, pp. 33–54. Cambridge University Press, Cambridge.
Helpman, E., M. Melitz, and Y. Rubinstein (2008). Estimating Trade Flows: Trading Partners and Trading Volumes. Quarterly Journal of Economics 123 (2), 441–
487.
Kancs, D. (2005). Can we use NEG models to predict migration flows? An example
of CEE accession countries. Migration Letters 2 (1), 32–63.
Kovak, B. (2011). Regional labor market effects of trade policy: Evidence from
Brazilian liberalization. Mimeo.
Krugman, P. (1991). Increasing returns and Economic Geography. Journal of Political Economy 99 (3), 483–499.
Levy, M. B. and W. J. Wadycki (1974). Education and the decision to migrate: An
econometric analysis of migration in Venezuela. Econometrica 42 (2), 377–88.
Lewis, W. A. (1954). Economic development with unlimited supplies of labour. The
Manchester School of Economic and Social Studies 22, 139–191.
Martin, W. and C. S. Pham (2008). Estimating the Gravity Equation when Zero
Trade Flows are Frequent. MPRA Paper 9453, University Library of Munich,
Germany.
Martinez-Zarzoso, I., F. Nowak-Lehmann, and S. Vollmer (2007). The Log of Gravity
Revisited. Center for European, Governance and Economic Development Research
(cege) Discussion Papers 64, University of Goettingen (Germany).
Menezes-Filho, N. A. and M.-A. Muendler (2007). Labor reallocation in response
to trade reform. Technical report.
Paillacar, R. (2006). Market potential and worker heterogeneity as determinants of Brazilian wages. Working paper, University of Paris 1. Available at
http://team.univ-paris1.fr/teamperso/paillacar/.
Pons, J., E. Paluzie, J. Silvestre, and D. Tirado (2007). Testing the New Economic Geography: Migrations and industrial agglomerations in Spain. Journal of
Regional Science 47 (2), 289–313.
35
Rauch, J. E. and V. Trindade (2002). Ethnic Chinese networks in international
trade. The Review of Economics and Statistics 84 (1), 116–130.
Redding, S. and P. K. Schott (2003). Distance, skill deepening and development:
Will peripheral countries ever get rich? Journal of Development Economics 72 (2),
515–541.
Redding, S. and A. Venables (2004). Economic geography and international inequality. Journal of International Economics 62 (1), 53–82.
Schwartz, A. (1973). Interpreting the effect of distance on migration. Journal of
Political Economy 81 (5), 1153–69.
Silva, J. M. C. S. and S. Tenreyro (2006). The Log of Gravity. Review of Economics
and Statistics 88 (4), 641–658.
Silva, J. M. C. S. and S. Tenreyro (2009). Comments on ”The Log of Gravity
Revisited”. Manuscript, London School of Economics.
Sorensen, T., P. Fishback, S. Allen, and S. Kantor (2007). Migration creation, diversion, and retention: New Deal grants and migration (1935-1940). IZA Discussion
Papers 3060, Institute for the Study of Labor (IZA).
Vasconcelos, J. and M. Oliveira (2006). Analise da Matriz por Atividade Economica
do Comercio Interestadual no Brasil. Working paper 1159, IPEA. Brazil.
36
Appendix A - Additional tables and figures
Figure A-1: Map of Brazil
Source: http://volunteerbrazil.com/photos%202008/map-of-brazil.gif
37
Table A-1: Industries
Comparative advantage Others
Agriculture
Food
Wood
Plastic & Non-metallics
Basic metals
Petroleum and gas
Mining
Textiles
Paper & Printing
Chemicals & Pharmaceutical
Electrical & electronics
Machinery
38
Appendix B - Data sources for market access calculations
In order to estimate sector-specific market access we need three sets of trade data: (1)
trade data between Brazilian states, which is drawn from Vasconcelos and Oliveira
(2006) who processed the information of the value-added tax provided by the National Council of Financial Policy (CONFAZ, Conselho Nacional de Politica Fazendaria) from the Ministry of Finance (Ministerio da Fazenda); (2) trade data between
Brazilian states and foreign countries (Secretaria de Comrcio Exterior, Ministry of
Trade); and (3) between foreign economies (BACI: Base pour lAnalyse du Commerce
International, CEPII). Moreover, using total sales across regions and industry from
the PIA database allows computing internal flows within regions by subtracting intra and international exports. These sets of data provide a complete and consistent
picture of all trade flows, defined at the 2 digit ISIC Revision 3 level (corresponding
to the Brazilian CNAE 2 digit industry classification).
We also need a variety of data on geography, infrastructures and regulations.
Distances, colonial links, languages, coordinates, GDP, areas and demographic densities are provided by CEPII (Centre dEtudes Prospectives et dInformations Internationales) and IBGE. The distance between states in the geodesic distance between
their respective capitals (computed in km using the coordinates).
39